Draft a customer reply with AI that a person reads before it goes

• Use a tool your employer has approved. Customer details must not go into a personal AI account.
• Give it your real policies and help articles to work from, so it doesn't rely on what it thinks a company like yours does.

Replace the words in square brackets with your own.

1. You
   Read the customer's message yourself first, and decide what they actually need.
2. AI
   Ask for a draft based only on your policy.

   PromptCopy

   If the message mentions harm, legal action or someone vulnerable, do not draft. Tell me to hand it to a person. If the product may be faulty, tell me to check the customer's legal rights.
   
   Otherwise, draft a reply to this customer. Use only the policy below. If the policy doesn't cover this, put a note to me above the draft and don't guess. Don't confirm what the customer says an agent told them. Do not promise refunds, dates or exceptions that the policy doesn't allow. Keep it under [words] words, and acknowledge the problem in the first line.
   
   Customer message (with personal details removed):
   [paste]
   
   Policy:
   [paste]
   
   After the draft, list which policy lines you used.

3. You
   Check it against the policy lines it listed, then change it to fit this customer.
4. You
   Send it yourself. If you corrected a mistake, report it so that the policy or the instructions can be fixed.

• Each promise, price and date is in the policy. A refusal doesn't override the customer's legal rights.
• It answers what this customer asked, not a similar question.
• It doesn't sound cheerful about something upsetting.
• Personal details belong to this customer, and nobody else's have slipped in.

• Complaints about harm, legal threats, vulnerable customers, or anything about health or money that needs a specialist. Hand these to a person from the first word.
• Letting replies go out with no person reading them. That is a different and riskier thing.